Cern Inspire MCP Server

What cern-inspire-mcp-server MCP server does

cern-inspire-mcp-server MCP server exposes INSPIRE-HEP research data through MCP tools. It covers literature, authors, experiments, collaborations, facilities, and HEPData measurement records. Agents can search for papers using INSPIRE query syntax or free text, retrieve a paper by record ID, arXiv ID, DOI, or supported URL, and inspect whether numerical data is available in HEPData.

The server also supports citation workflows. It can return BibTeX or LaTeX \\bibitem entries for papers selected by a literature query, and it can calculate h-index values, citation totals, averages, citation buckets, and citations by year for an author or literature query. The cern_inspire_list_reference tool explains supported query syntax, identifier formats, filters, citation buckets, and HEPData DOI handling.

How it works

The server queries INSPIRE-HEP and its HEPData index on behalf of an MCP client. Literature searches accept sorting, document-type and subject filters, year ranges, result sizes, and pages. Search results include identifiers, titles, first authors, dates, citation counts, arXiv IDs, DOIs, publication details, and an abstract excerpt.

A paper lookup returns a fuller record and can report a linked HEPData record, its DOI, latest version, table count, and URL. The literature resource inspire://literature/{recid} provides the same paper dossier in JSON for clients that expose MCP resources.

Citation summaries accept exactly one author identifier or literature query. Author targets must use a BAI, ORCID, INSPIRE ID, or author record ID; names should first be resolved with author search. Optional year, subject, document-type, and self-citation filters affect the summary.

Setup and configuration

cern-inspire-mcp-server MCP server can run as a stdio process or as a local Streamable HTTP server. The repository also publishes a public hosted MCP endpoint at https://cern-inspire.caseyjhand.com/mcp. The README provides installation links for Claude Desktop, Cursor, and VS Code, using the published npm package @cyanheads/cern-inspire-mcp-server.

A package-based setup can launch the server through npx with the package name. The repository also publishes a Docker image and identifies Bun as part of its runtime tooling. No required environment variables are documented in the provided material.

Tools and capabilities

The eight tools are:

• cern_inspire_search_literature searches and pages through INSPIRE papers.
• cern_inspire_get_paper retrieves a paper record and HEPData availability.
• cern_inspire_export_citations exports BibTeX or LaTeX citation entries.
• cern_inspire_search_authors finds profiles by name or supported identifier.
• cern_inspire_get_citation_summary reports author or query-level citation metrics.
• cern_inspire_search_experiments finds experiments, collaborations, accelerators, and facilities.
• cern_inspire_search_hepdata searches measurement records by physics-related fields.
• cern_inspire_list_reference documents query and identifier conventions.

Limitations and notes

Literature search and citation export can reach only the first 10,000 results for a query. Search pages are limited to 100 results, while citation export pages are limited to 50. Author and experiment searches return at most 25 results per request.

Citation summaries have additional constraints. Per-year citation data may be omitted when year filters or self-citation exclusion are used, when INSPIRE does not return the series, or for very large queries. An author name alone is not a valid citation-summary target; resolve it to an identifier first. Paper records can also represent merged records or distinguish a HEPData record from a paper when a supplied identifier points to HEPData rather than INSPIRE literature.